Extreme precipitation and floods are among the most common and impactful disasters in the US and other parts of the world. Advancing the process-level understanding and forecasting capability of these events has become increasingly important. We have been using Earth system modeling tools and Earth observations to study recent flood events in different US regions (e.g., an Ellicott City extreme flood in May 2018; and the Oroville Dam Crisis in February 2017 following atmospheric river storms). Here, we discuss results from multiple model simulations using different configurations that help address the following questions regarding the partitioning of water balance and soil moisture-precipitation relationships: 1) How does the antecedent (i.e., at the model initialization times prior to the floods) soil moisture as well as other land surface fields (e.g., snow water equivalent and depth) impact the extreme precipitation prediction, and to what extent precipitation forecasts can be improved via adjusting the model's land initial conditions? 2) How does the accuracy of the predicted precipitation, affect the accuracy of the modeled soil moisture changes and flood forecasts during these events? 3) How does the forecast lead time as well as the chosen atmospheric forcing and physics schemes affect the model performance and the findings from 1) and 2), and why? In addition, we introduce planned modeling experiments in support of a new GEWEX/SoilWat (https://www.gewex.org/panels/global-landatmosphere-system-study-panel/glass-projects/gewex-soilwat-initiative) initiative “soil-cloud cascade”, that aims to investigate in detail the impact of soil properties, in particular soil texture and related hydraulic and thermal properties, on soil moisture, surface fluxes, clouds and precipitation. The interconnections between extreme floods and other hazards such as wildfires and pollution, as well as their co-impacts on various communities, will be integrated into the discussions.